Conventional uses of neural networks are typically limited by an amount of computational resources available for implementation of the neural network. Accordingly, in order to employ a conventional neural network to process an image, the image is downsized in conventional techniques such that the amount of image data to be processed does not overwhelm the available computational resources. Consequently, a large amount of information may be lost in downsizing the image, which could make these conventional neural network processing techniques ill-suited for some image processing operations.